Technical Aspects of Mammography Image Acquisition

Mammograms are obtained on specially designed, dedicated x-ray machines using either x-ray film and paired fluorescent screens or digital detectors to capture the image. All mammography units are comprised of a rotating anode x-ray tube with matched filtration for soft-tissue imaging, a breast compression plate, a moving grid, an x-ray image receptor, and an automatic exposure control (AEC) device that can be placed under or detect the densest portion of the breast, all mounted on a rotating C-arm (Fig. 1-1). A technologist compresses the patient’s breast between the image receptor and compression plate for a few seconds during each exposure. Breast compression is important because it spreads normal fibroglandular tissues so that cancers, which have similar attenuation properties to fibroglandular tissues, can be better seen. Breast compression also decreases breast thickness, thereby decreasing exposure time, radiation dose to the breast, and the potential for image blurring as a result of patient motion and unsharpness.

Women worry about breast pain from breast compression and about the radiation dose from mammography. Breast pain during compression varies among individuals and can be decreased by obtaining the mammogram 7 to 10 days after the onset of menses, when the breasts are least painful. Breast pain can also be minimized by taking oral analgesics such as acetaminophen before the mammogram or by using appropriately designed foam pads that cushion the breast without adversely affecting image quality or increasing breast dose.

Current mammography delivers a low dose of radiation to the breast. The most radiosensitive tissues in the breast are the fibroglandular tissues. The best measure of breast dose is mean glandular dose, or the average absorbed dose of ionizing radiation to the radiosensitive fibroglandular tissues. The mean glandular dose received by the average woman is approximately 2 mGy (0.2 rad) per exposure, or 4 mGy (0.4 rad) for a typical two-view examination. Radiation doses to thinner compressed breasts are substantially lower than doses to thicker breasts.

The main radiation risk from mammography is the possible induction of breast cancer 5 to 30 years after exposure. The estimated risk of inducing breast cancer is linearly proportional to the radiation dose and inversely related to age at exposure. The lifetime risk of inducing a fatal breast cancer as a result of two-view mammography in women age 45 at exposure is estimated to be about 1 in 100,000. For a woman age 65 at exposure, the risk is less than 0.3 in 100,000. The benefit of screening mammography is the detection of breast cancer before it is clinically apparent. The likelihood of an invasive or in-situ cancer being present in a woman at age 45 is about 1 in 500. The likelihood that the cancer would be fatal in the absence of mammography screening is about 1 in 4, and the likelihood that screening mammography will convey a mortality benefit is 15% (RCT estimates) to 40% (service screening estimates). Hence, the likelihood of screening mammography saving a woman’s life at this age is about 1 in 5000 to 1 in 13,000, yielding a benefit-to-risk ratio of 8 : 1 to 20 : 1. For a woman age 65 at screening, the likelihood of a mortality benefit from mammography is about 1 in 2000 to 1 in 4000 (assuming a 25% to 50% mortality benefit), yielding a benefit-to-risk ratio of approximately 90 : 1 to 180 : 1. Screening mammography is only effective when regular periodic examination is performed.

The generator for a mammography system provides power to the x-ray tube. The peak kilovoltage (kVp) of mammography systems is lower than that of conventional x-ray systems, because it is desirable to use softer x-ray beams to increase both soft-tissue contrast and the absorption of x-rays in the cassette phosphor (absorption efficiency), especially for screen-film mammography (SFM). Typical kVp values for mammography are 24 to 32 kVp for molybdenum targets, 26 to 35 kVp for rhodium or tungsten targets. A key feature of mammography generators is the electron beam current (milliampere [mA]) rating of the system. The higher the mA rating, the shorter the exposure time for total tube output (milliampere second [mAs]). A compressed breast of average thickness (5 cm) requires about 150 mAs at 26 kVp to achieve proper film densities in SFM. If the tube rating is 100 mA (typical of the larger focal spots used for nonmagnification mammography), the exposure time would be 1.5 seconds. A higher-output system with 150-mA output would cut the exposure time to 1 second for the same compressed breast thickness and kVp setting. Because of the wide range of breast thicknesses, exposures require mAs values ranging from 10 to several hundred mAs. Specifications for generators are listed in Box 1-3.

The most commonly used anode/filter combination is Mo/Mo: a molybdenum (Mo) anode (or target) and a Mo filter (25–30 microns thick), especially for thinner compressed breasts (<5 cm thick). Most current manufacturers also offer a rhodium (Rh) filter, to be used with the Mo target (Rh/Mo), to produce a slightly more penetrating (harder) x-ray beam for use with thicker breasts. Some manufacturers offer other target materials, such as Rh/Rh: a rhodium target paired with a rhodium filter, or tungsten (W), which is paired with a rhodium filter (W/Rh) or aluminum (Al) filter (W/Al). These anode/filter combinations are designed for thicker (>5 cm) and denser breasts. Typically, higher kVp settings are also used with these alternative target/filter combinations to result in a harder x-ray beam for thicker breasts, because fewer x-rays are attenuated with a harder x-ray beam (Box 1-4). One of the best parameters to measure the hardness or penetrating capability of an x-ray beam is the half-value layer (HVL), which represents the thickness of aluminum that reduces the exposure by one half. The harder the x-ray beam, the higher the HVL. The typical HVL for mammography is 0.3 to 0.5 mm of Al. The Food and Drug Administration (FDA) requires that the HVL for mammography cannot be less than kVp/100 ± 0.03 (in mm of Al), so that the x-ray beam is not too soft. For example, at 28 kVp, the HVL cannot be less than 0.31 mm of Al. There is also an upper limit on the half-value layer that depends on the target-filter combination. For the upper limit of Mo/Mo, the HVL must be less than kVp/100 + 0.12 (in mm of Al); so for 28 kVp, the HVL must be less than 0.4 mm of Al.

The size of the larger mammography focal spot used for standard, contact (i.e., nonmagnification) mammography is typically 0.3 mm. Magnification mammography requires a smaller focal spot, about 0.1 mm, to reduce penumbra (geometric blurring of structures in the breast produced due to the breast being closer to the x-ray source and farther from the image receptor to produce greater “geometric” magnification). The effect of focal spot size on resolution in the breast is tested by placing a line pair pattern in the location of the breast, at a specific distance (4.5 cm) from the breast support surface. For SFM, the larger mammography focal spot used for standard, contact mammography should produce an image that resolves at least 11 line pairs/millimeter (11 lp/mm) when the lines of the test pattern run in the direction perpendicular to the length of the focal spot (this measures the blurring effect of the length of the focal spot) and at least 13 lp/mm when the lines run parallel to the focal spot (measuring the blurring effect of the width of the focal spot). Thus, although the SFM image receptor can resolve 18 to 21 lp/mm, the geometry of the breast in contact mammography and the finite-sized larger focal spot reduce the limiting spatial resolution of the system to 11 to 15 lp/mm in the breast. The limiting spatial resolution of digital mammography systems is less (5–10 lp/mm), due to pixelization of the image by the digital image receptor. In digital, a “line” is 1 pixel width, and a line pair is 2 pixels. For example, for a digital detector with 100 micron (0.1 mm) pixel size or pitch (the center-to-center distance between adjacent pixels), a line pair consists of 2 pixels or 200 microns (0.2 mm). Therefore, one can fit 5 line pairs (at 0.2 mm each) into a 1 mm length, or the detector has a limiting spatial resolution of 5 lp/mm. By similar reasoning, a digital detector with 50 micron pixels has a limiting spatial resolution of 10 lp/mm.

The x-ray tube and image receptor are mounted on opposite ends of a rotating C-arm to obtain mammograms in almost any projection. The source-to-image receptor distance (SID) for mammography units must be at least 55 cm for contact mammography. Most systems have SIDs of 65 to 70 cm.

Geometric magnification is achieved by moving the breast farther from the image receptor (closer to the x-ray tube) and switching to a small focal spot (Fig. 1-2). Placing the breast halfway between the focal spot and the image receptor (as in Fig. 1-2B) would magnify the breast by a factor of 2.0 from its actual size to the image size because of the divergence of the x-ray beam. The MQSA requires that mammography units with magnification capabilities must provide at least one fixed magnification factor of between 1.4 and 2.0 (Table 1-1). Geometric magnification makes small, high-contrast structures such as microcalcifications more visible by making them larger relative to the noise pattern in the image (increasing their signal-to-noise ratio [SNR]). Optically or electronically magnifying a contact image, as would be done with a magnifier on SFM or using a zoom factor greater than 1 on a digital mammogram, does not increase the SNR of the object relative to the background, because both are increased in size equally. To avoid excess blurring of the image with geometric magnification, it is important to use a sufficiently small focal spot (usually 0.1 mm nominal size) and not too large a magnification factor (2.0 or less). When the small focal spot is selected for geometric magnification, the x-ray tube output is decreased by a factor of 3 to 4 (to 25–40 mA) compared to that from a large focal spot (80–150 mA). This can extend imaging times for magnification mammography, even though the grid is removed in magnification mammography. The air gap between the breast and image receptor provides adequate scatter rejection in magnification mammography without the use of an antiscatter grid.

Table 1-1 Mammography Focal Spot Sizes and Source-to-Image Distances

Mammography Type	Nominal Focal Spot Size (mm)	Source-to-Image Distance (cm)
Contact film-screen	0.3	≥55
Magnification	0.1	≥55

The Mammography Quality Standards Act requires magnification factors between 1.4 and 2.0 for systems designed to perform magnification mammography.

Collimators control the size and shape of the x-ray beam to decrease patient exposure to tissues beyond the compressed breast and image receptor. In mammography, the x-ray beam is collimated to a rectangular field to match the image receptor rather than the breast contour, because x-rays striking the image receptor outside the breast do not contribute to breast dose. By federal regulation, the x-ray field cannot extend beyond the chest wall of the image receptor by more than 2% of the SID. So, for a 60-cm SID unit, the x-ray beam can extend beyond the chest wall edge of the image receptor by no more than 1.2 cm.

The compression plate and image receptor assembly hold the breast motionless during the exposure, decreasing the breast thickness and providing tight compression, better separating fibroglandular elements in the breast (Fig. 1-3). The compression plate has a posterior lip that is more than 3 cm high and usually is oriented at 90 degrees to the plane of the compression plate at the chest wall. This lip keeps chest wall structures from superimposing and obscuring posterior breast tissue in the image. The compression plate must be able to compress the breast for up to 1 minute with a compression force of 25 to 45 pounds. The compression plate can be advanced by a foot-controlled motorized device and adjusted more finely with hand controls.

Lost contrast Film turns brown Inadequate rinsing of fixer Motion artifact

For positioning, the technologist tailors the mammogram to the individual woman’s body habitus to get the best image. The breast is relatively fixed in its medial borders near the sternum and the upper breast, whereas the lower and outer portions of the breast are more mobile. The technologist takes advantage of the mobile lower outer breast to obtain as much breast tissue on the mammogram as possible. The two views obtained for screening mammography are the craniocaudal (CC) and mediolateral oblique (MLO) projections. The names for the mammographic views and abbreviations are based on the ACR Breast Imaging Reporting and Data System (ACR BI-RADS®), a lexicon system developed by experts for standard mammographic terminology. The first word in the mammographic view indicates the location of the x-ray tube, and the second word indicates the location of the image receptor. Thus, a CC view would be taken with the x-ray tube pointing at the breast from the head (cranial) down through the breast to the image receptor in a more caudal position.

To pass ACR accreditation clinical image review, the MLO mammogram must show most of the breast tissue in one projection, with portions of the upper inner and lower inner quadrants partially excluded (Fig. 1-4). Clinical evaluation of the MLO view should show fat posterior to the fibroglandular tissue and a large portion of the pectoralis muscle, which should be concave and extend inferior to the posterior nipple line (PNL). The PNL describes an imaginary line drawn from the nipple to the pectoralis muscle or film edge and perpendicular to the pectoralis muscle. The PNL should intersect the pectoralis muscle in the MLO view in more than 80% of women. Although the technologist tries to avoid producing skin folds on the film when possible, they are seen occasionally but do not usually cause problems for the radiologist reading the film. The MLO view should show adequate compression, exposure, contrast, and an open inframammary fold, in which both the lower portion of the breast and a portion of the upper abdominal wall should be seen.

To pass ACR accreditation clinical image review, the CC view should include the medial posterior portions of the breast without sacrificing the outer portions (Figs. 1-5 and 1-6). With proper positioning technique, the technologist should be able to include the medial portion of the breast without rotating the patient medially by lifting the lower medial breast tissue onto the image receptor. The pectoralis muscle should be seen when possible on the CC view. On the CC view, the PNL extends from the nipple to the pectoralis muscle or the edge of the film, whichever comes first, perpendicular to the pectoralis muscle or film edge. For a given breast, the length of the PNL on the CC view should be within 1 cm of its length on the MLO view.

Clinical images are evaluated on positioning, compression, contrast, proper exposure, random noise (radiographic mottle or quantum mottle produced by varying numbers of x-rays contributing to the image in different locations, even with a uniform object), sharpness, and artifacts (or structured noise). Imaging on a phantom is helpful in evaluating most of these factors, except for positioning and compression (Fig. 1-7). Adequate exposure (to achieve adequate film OD) and adequate contrast (OD difference) are important to ensure detection of subtle abnormalities (Fig. 1-8). Artifacts seen on clinical images include processing artifacts (roller marks, wet pressure marks, guide shoe marks), white specklike artifacts from dust or lint between the fluorescent screen and film emulsion, grid lines from incomplete grid motion, motion artifacts from patient movement (made more likely by longer exposure times), skin folds from positioning, tree static caused by static electricity from low humidity in the dark room, or film handling artifacts (fingerprints, crimp marks, or pressure marks) (Figs. 1-9 to 1-12).

Film labeling is important (Box 1-6) because proper labeling ensures accurate facility, patient, laterality, and projection identification. Guidelines from the ACR Mammography Accreditation Program for mammogram labeling state that an identification label on the mammogram should specify the patient’s first and last name, unique identification number, facility name and address, date, view and laterality, an Arabic number indicating the cassette used, and the technologist’s initials. The laterality and projection marker should be placed near the axilla on all screen-film views.

Digital Mammography Image Acquisition

In digital mammography, the image is obtained in the same manner as in screen-film mammography, using a compression plate and an x-ray tube, with the screen-film cassette replaced by a digital detector (Figs. 1-13 and 1-14). Digital image acquisition has several potential advantages in terms of image availability, image processing, and CAD (Fig. 1-15). One advantage is elimination of the film processor, which eliminates artifacts and image noise added by processing films.

Digital mammography uses indirect or direct digital detectors. Indirect digital detectors use a fluorescent screen made of materials such as cesium iodide (CsI) to convert each absorbed x-ray to hundreds of visible light photons. Behind the fluorescent material, light-sensitive detector arrays made of materials such as amorphous silicon diodes or charge-coupled devices (CCDs) measure the produced light pixel by pixel. The weak electronic signal measured in each pixel is amplified and sent through an analog-to-digital converter, enabling computer storage of each pixel’s measured detector signal.

Direct digital detectors use detector elements that capture and count x-rays directly, although amplification and analog-to-digital conversion are still applied. Another method to produce digital mammograms involves amorphous selenium. An amorphous selenium plate is an excellent absorber of x-rays and an excellent capacitor, storing the charge created by ionization when x-rays are absorbed. After exposure, an electronic device is used to read out the charge distribution on the selenium plate, which is in proportion to local exposure. This can be done by scanning the selenium plate with a laser beam or by placing a silicon diode array in contact with one side of the plate, with bias voltage applied, to read out the stored charge. Each of these methods allows production of high-resolution digital images.

Another approach to full-field digital mammography (FFDM) is computed radiography (CR), which uses a photostimulable phosphor composed of barium fluorobromide doped with europium (BaFBr:Eu). CR uses the same dedicated mammography units as screen-film, replacing the screen-film cassettes and film processor with CR cassettes (in sizes of 18 × 24 cm and 24 × 30 cm) and a CR processor. The phosphor plate within the CR cassette is used to absorb x-rays just as the screen in a screen-film cassette does. Rather than emitting light immediately after exposure (through fluorescence), x-ray absorption in the phosphor causes electrons within the phosphor crystals to be promoted to higher energy levels (through photostimulation). The plate is removed from the cassette in the CR processor and a red laser light scans the phosphor plate point by point, releasing electrons and stimulating emission of a higher energy (blue) light in proportion to x-ray exposure. In conventional x-ray systems, CR phosphor plates have an opaque backing and are read from only one side. In the one FDA-approved CR system for mammography (Fuji 5000D CR, Fujifilm Medical Systems USA), the CR cassette base is transparent and light emitted from the plate during laser scanning is read from both sides to increase reading efficiency.

No matter which digital detector is used, its job is to measure the quantity of x-rays passing through the breast, compression plate, grid (in most cases), and breast holder. The signal measured in each pixel is determined by the total attenuation in the breast along a given ray.

The choice of an analog-to-digital converter determines how many bits of memory will be used to store the signal for each pixel; the more bits per pixel, the more dynamic range for the image, but at higher storage cost. Specifically, if 12 bits per pixel are used, 2¹² or 4096 signal values can be stored. If 14 bits per pixel are used, 2¹⁴ or 16,384 signal values can be stored. Usually 12- to 14-bits storage per pixel is used. In either case, 2 bytes per pixel are required (8 bits = 1 byte) to store the image. For example, the GE Senographe 2000D and DS digital detectors have 1920 × 2304 pixel arrays, or 4.4 million pixels, requiring 8.8 million bytes (8.8 megabytes, MB) of storage per image. Other FFDM systems require up to 52 MB of storage per image.

Screen-film image receptors used for mammography have a line-pair resolution of 18 to 21 lp/mm. To equal this spatial resolution, a digital detector would require 25-micron pixels, which would yield noisier images and pose a storage issue due to the large data sets required to store those images. FFDM systems have spatial resolutions ranging from 5 lp/mm (for 100-micron pixels) to 10 lp/mm (for 50-micron pixels). In digital mammography systems, it is the size of the pixels, or more correctly their center-to-center distance (pitch), that determines (and limits) the spatial resolution of the imaging chain.

The lower limiting spatial resolution of FFDM systems compared to film is offset by the increased contrast resolution of FFDM systems. Unlike SFM, in which the image cannot be manipulated after exposure and processing, FFDM images can be optimized after image capture by image postprocessing and adjustment of image display. For fixed digital detectors, such as CsI and silicon diode arrays (used by GE) and selenium and amorphous silicon diode arrays (used by Hologic and Siemens), one image processing step that can minimize image noise and structured artifacts is flat-field correction, or “gain correction” of each acquired digital image. This is done by making and storing a sensitivity map of the digital detector and using that map to correct all exposures. Typically, slot-scanning devices (such as the older SenoScan digital system, Fischer Medical Systems) and CR systems do not have the ability to perform flat-field correction of digital images. Beyond this, all digital systems have the ability to process the acquired digital image to minimize or eliminate the signal difference that results from the roll-off in thickness of the breast toward the skin line (thickness equalization); some add processing to help enhance the appearance of microcalcifications (e.g., GE Premium View and FineView). The window width and window level for all digital images viewed with soft copy display on review workstations can be adjusted, changing the contrast and brightness of the images, respectively, as well as digitally magnifying images.

Another important difference between SFM and FFDM is that screen-film images have a linear relationship between the logarithm of x-ray exposure and film OD only in the central portion of the characteristic curve. In FFDM, there is a linear relationship between x-ray exposure and signal over the entire dynamic range of the detector. Thus, digital images do not suffer contrast loss in underexposed or overexposed areas of the mammogram (as long as detector saturation does not occur), and instead show similar contrast over the full dynamic range of signals. FFDM also eliminates the variability and noise added by film processing.

In terms of breast dose, FFDM has a mean glandular dose lower than, or comparable to, the radiation dose of SFM. Recent results from the American College of Radiology Imaging Network (ACRIN) Digital Mammographic Imaging Screening Trial (DMIST) found the average single-view mean glandular dose for FFDM to be 1.86 mGy, 22% lower than the average SFM mean glandular dose of 2.37 mGy (Hendrick et al., 2010). Specific manufacturers, especially those using slot-scanning techniques, produce lower doses than SFM. Slot-scanning systems, such as the Fischer SenoScan, have a narrow slot of detector elements that are scanned under the breast in synchronization with a narrow fan beam of x-rays swept across the breast. This design, although more technically difficult to implement, has the advantage of eliminating the need for a grid to reduce scattered radiation. Scatter is partially eliminated by the narrow slot itself. The absence of a grid reduces the amount of radiation to the breast needed to get the same SNR in the detector. Some full area detectors with AEC systems have also demonstrated lower breast doses compared to SFM, especially for thicker breasts.

Once captured and processed, the image data are transferred to a reading station for interpretion on high-resolution (2048 × 2560, or 5 Mpixel) monitors or printed on films by laser imagers (with approximately 40-micron spot sizes) for interpretation of hardcopy images on film viewboxes or alternators. Digital data can be stored on optical disks, magnetic tapes, Picture Archiving and Communication Systems (PACS), or on CD-ROMs for later retrieval.

The MQSA states that FFDM images must be made available to patients as hardcopy films, as needed, which means the facility must have access to an FDA-approved laser printer for mammography that can reproduce the gray scale and spatial resolution of FFDM films. The images may also be given to the patient on a CD, if acceptable to the patient.

A number of studies have evaluated the performance of FFDM compared to SFM for screening asymptomatic women for breast cancer. Early studies showed comparable or slightly worse results (but not statistically significant differences) for receiver operating characteristic (ROC) curve area and sensitivity (Lewin et al.) or cancer detection rate (Skaane and Skjennald) of FFDM compared to SFM. Larger studies, however, showed some benefits of FFDM compared to SFM. The ACR Imaging Network Digital Mammographic Imaging Screening Trial (ACRIN DMIST) paired study (Pisano and colleagues), showed no difference overall, but found that FFDM had statistically significantly higher ROC curve areas than SFM for women under age 50, for premenopausal and perimenopausal women, and for women with denser breasts (BI-RADS® density categories 3 and 4). The Oslo II trial (Skaane et al.) showed that digital mammography had a significantly higher cancer detection rate (5.9 cancers per 1000 women screened) than SFM (3.8 cancers per 1000 women screened).

As of May 2010, about two thirds (65.4%) of the mammography units in the United States were digital mammography systems. The FDA-approved manufacturers for digital mammography and their unit properties are listed in Table 1-3.

Quality Assurance in Mammography and the Mammography Quality Standards Act

MQSA, a federal law regarding mammography that is enforced by the FDA, stipulates that all institutions performing mammography must be certified by the FDA. A prerequisite to FDA certification is accreditation to perform mammography by an FDA-approved accrediting body, such as the ACR or an FDA-approved state accrediting body. Arkansas, Iowa, and Texas are approved to accredit mammography facilities in their own states. MQSA regulations are listed in the Federal Register. To update facilities on the latest regulation changes and updates, the FDA maintains a Web site on MQSA (www.fda.gov/cdrh/mammography/) that includes a section to guide users who have questions on MQSA compliance (www.fda.gov/cdrh/mammography/guidance-rev.html).

MQSA certification involves an initial application to the FDA and FDA approval to perform mammography, continuous documentation of compliance, and yearly facility inspection by an MQSA or state inspector. Noncompliance with regulations may result in FDA citations, with time limits on deficiency corrections. Serious noncompliance issues may result in facility closure. Falsification of data submitted to the FDA can result in monetary fines and jail terms.

MQSA equipment requirements for mammography are summarized in Box 1-7. MQSA qualification requirements for radiologists, technologists, and medical physicists are outlined in Boxes 1-8 to 1-10.

One radiologist at each facility must be designated the supervising interpreting physician to oversee the facility’s quality assurance (QA) program (Boxes 1-11 and 1-12). The supervising physician oversees assessment of mammography outcomes to evaluate the accuracy of interpretation. The facility must have a method for recording outcomes on interpretation of all abnormal mammographic findings and tallying these interpretations for each individual physician and for the group as a whole, providing feedback to each radiologist on a yearly basis (Box 1-13). A portion of the medical audit includes review of the pathology in cases recommended for biopsy.

One radiologic technologist designated the QC technologist oversees the quality control (QC) tasks outlined in Table 1-4, which specifies the minimum frequency of each QC test and action limits for test performance. One important test performed by the QC technologist and reviewed by the interpreting physician is evaluation of the mammography phantom image; this test is performed at least weekly and evaluates the entire imaging system. The phantom consists of fibers, speck clusters, and masses of various sizes imbedded in a uniform phantom material. The technologist takes a phantom radiograph using the site’s clinical technique for a 4.5-cm thick compressed breast, the radiograph is processed on the site’s film processor, and the image is evaluated for the number of objects seen in each category. To pass accreditation and meet MQSA requirements, the phantom should show a minimum of four fibers, three speck groups, and three masses (Box 1-14). The phantom image should also be free of significant artifacts. These and other tests are used to evaluate the entire imaging system.

Table 1-4 Technologist Quality Control Tests for Screen-Film Mammography

Periodicity	Quality Control Test	Desired Result
Daily	Darkroom cleanliness	No dust artifacts
Daily	Processor quality control	Density difference and mid-density changes not to exceed control limits of ±0.15
Weekly	Screen cleanliness	No dust artifacts on films
Weekly	View box cleanliness	No marks on panels, uniform lighting
Weekly	Phantom image evaluation	Film density >1.4 with control limits of ±0.20 Densities do not vary over time or between units Minimum test objects seen: 4 largest fibers, 3 largest speck groups, 3 largest masses

Monthly Visual checklist Each item on checklist present and functioning properly Quarterly Repeat analysis Quarterly Analysis of fixer retention Residual sodium thiosulfate (hypo) ≤0.05 µg/cm³ Semiannually Darkroom fog Fog ≤0.05 optical density difference for 2-min exposure in darkroom Semiannually Screen-film contact Large areas (>1 cm) of poor contact unacceptable Semiannually Compression

From Hendrick et al: Mammography quality control manual. Reston, VA, 1999, American College of Radiology, p. 119.

The medical physicist surveys the equipment just after installation, after important major equipment repairs or upgrades, and annually, performing the QC tests outlined in Box 1-15. The medical physicist’s survey report is an important component of the QA program and is reviewed by the supervising physician to ensure high-quality mammography. The facility is responsible for correcting deficiencies pointed out by the site medical physicist.

Each year, the mammography facility is inspected by state or federal inspectors who evaluate compliance with MQSA regulations. Site QA records and site personnel qualifications are routinely checked by the MQSA inspector. Correction of deficiencies specified in the medical physicist’s report is an important item checked by MQSA inspectors. Noncompliance with MQSA requirements may result in warnings requiring corrective actions or, in extreme cases, facility closure.

Computer-Aided Detection

Radiologists are trained to detect early, subtle signs of breast cancer, such as pleomorphic calcifications and spiculated masses on mammograms. CAD systems use algorithms to review mammograms for bright clustered specks and converging lines, which represent pleomorphic calcifications and spiculated masses, respectively. These programs were developed to help radiologists search for signs of cancer against the complex background of dense breast tissue and fat.

Some facilities use CAD as a second reader. Double reading in screening mammography involves two observers reviewing the same mammograms to increase detection of cancer, decrease the false-negative rate or, in some facilities, decrease the false-positive recall rate by using a consensus. Studies have shown that double reading, depending on its implementation, increases the rate of detection of cancer by 5% to 15%. However, the expense and logistic problems of implementing a second interpreting radiologist limit the practice of double reading of mammography in clinical practice in the United States.

Mammography data used for CAD algorithms are obtained digitally from FFDM units or are digitized from screen-film mammograms. The digital or digitized mammograms undergo analysis by computer schemes, which mark potential abnormal findings on a low-resolution paper print or monitor image (Fig. 1-16). For FFDM, CAD marks potential abnormalities directly on the image displayed on the workstation monitor. The radiologist interprets and analyzes the marked findings, and each finding is dismissed as insignificant or recalled for further workup (Fig. 1-17).

CAD algorithms detect microcalcifications, masses, and parenchymal distortions on images using computer schemes derived from large numbers of mammograms in which biopsy results are known. The computer scheme’s ability to mark true cancers is optimized by reviewing the “true positive” and “false positive” marks on the training set of mammograms. These optimized algorithms are later tested on both known subtle and obvious cancers. Using the optimized schemes, commercial CAD systems mark abnormalities that represent cancers (“true positive” marks, a measure of CAD sensitivity), and findings that do not represent cancer or where no known cancer has occurred (“false positive” marks, a measure of CAD specificity) (Fig. 1-18). Because detection of masses or calcifications by the CAD scheme is directly affected by image quality, good-quality mammograms are required to obtain good CAD output. Mammograms of suboptimal quality will result in poor CAD output. CAD output also can be affected by the type and reproducibility of the digitizer if the data is from digitized SFM. Thus, it is essential to have high-quality mammograms since CAD cannot overcome poor image quality.

The FDA has approved CAD systems for breast cancer detection in both screening and diagnostic mammography using both screen-film and digital mammography, including the Fuji CR digital system.

A retrospective study of breast cancers found on mammography by Warren-Burhenne and colleagues determined that a CAD program marked 77% (89/115) of screening-detected breast cancers. Birdwell and colleagues reviewed “negative” mammograms obtained the year before the diagnosis of 115 screen-detected cancers in 110 patients. They reported that a CAD program marked reader-missed findings in 77% (88/115) of false-negative mammograms. Specifically, CAD marked 86% (30/35) of missed calcifications and 73% (58/80) of missed masses.

Freer and Ulissey reported that in a prospective community breast center study of 12,860 women undergoing screening mammography, CAD increased their cancer detection rate by 19.5%. Radiologists detected 41 of 49 cancers and missed 8 cancers found by the CAD system (7 of 8 were calcifications). CAD detected 40 of the 49 cancers, but it did not mark 9 radiologist-detected masses that were proven to be cancers.

It is important to note that CAD systems did not diagnose all cancers, nor should they be used as the only evaluator of screening mammograms. In Freer and Ulissey’s study, radiologists initially made a decision about the mammogram and then used CAD and re-reviewed the marked mammogram. The radiologist’s decision to recall a potential abnormality could not be changed by failure of the CAD system to mark the potential finding. Findings marked by CAD could be recalled even if the finding was not initially detected by the radiologist but was judged to be abnormal in retrospect. This means that radiologists should read the mammogram first so they are not influenced by CAD marks initially because not all cancers are marked by CAD.

CAD marks have low specificity inasmuch as approximately 97.6% of the CAD marks were dismissed by the interpreting radiologists in the study by Freer and Ulissey. The radiologist had identified almost all of the 2.4% of CAD-marked findings that were selected for recall, which means that high-sensitivity CAD systems will mark significant potential findings as well as numerous insignificant findings, thus identifying tumors but marking a number of normal findings that must be dismissed. Accordingly, it is expected that many insignificant findings will be marked by the CAD system, most of which can be dismissed readily, and yet the radiologists’ attention will still be drawn to overlooked suspicious areas.

Other CAD studies have shown somewhat less positive results. Gur and colleagues assessed changes in screening mammography recall rate and cancer detection rate after the introduction of a CAD system into a single academic radiology practice. Based on 56,432 cases interpreted without CAD and 59,139 cases interpreted with CAD by 24 radiologists, recall rates were identical without and with CAD (11.39% versus 11.40%, respectively; p = 0.96), as were breast cancer detection rates without and with CAD (3.49 versus 3.55 per 1000 women screened, respectively; p = 0.68). Feig and colleagues used Gur’s data to point out that lower-volume readers benefited from CAD by having a 19.7% higher cancer detection rate, but at the price of a 14% increase in recall rate, from 10.5% to 12%.

Fenton and colleagues conducted a retrospective study comparing SFM without and with CAD in early implementation (2–25 months). They showed that adding CAD led to a nonsignificant increase in sensitivity (from 80.4% without CAD to 84% with CAD; p = 0.32), a significant decrease in specificity (from 90.2% without CAD to 87.2% with CAD; p < 0.001), and a significant decrease in accuracy (area under the ROC curve decreased from 0.919 without CAD to 0.871 with CAD; p = 0.005).

A study by Gromet compared CAD-aided reading of screening mammograms to double-reading without CAD. He found that CAD-aided reading had a nonsignificantly higher sensitivity than double-reading (90.4% versus 88.0%), with a significantly lower recall rate (10.6% versus 11.9%, respectively; p < 0.0001).

CAD programs have the potential to increase detection of cancer, particularly for readers with less experience or lower reader volumes, perhaps at the price of somewhat lower specificity and slightly longer interpretation times. In the end, however, it is the radiologist’s knowledge and interpretive skills that have an impact on cancer detection, whether CAD is used or not.

Conclusion

Mammography acquisition is affected by a number of variables, including factors affecting image quality, such as the x-ray equipment, processing technique, technologist positioning, breast compression, patient differences such as breast density, lesion type, and the skills of the interpreting radiologist. It is important for radiologists to understand equipment requirements and the effect of imaging parameters on image quality; in addition, practitioners should be able to solve imaging problems that occur in everyday clinical practice. MQSA regulations were put into effect to mandate many of the factors that are known to affect image quality and to improve the quality of mammography. Every radiologist who performs breast imaging should understand MQSA requirements and be able to supervise a high-quality mammography practice, working toward improved technical quality and interpretive skills based on follow-up and feedback from sound quality assurance practices.

Key Elements